Design And Implementation Of Match-line Voltage Swing Self-adjustable TCAM

Ternary Content-Addressable Memory(TCAM)is a high-speed parallel hardware searchengine which provides a hardware-based solution for many application scenarios that require high-speed search operations.These applications put higher and lower power consumption and search speed requirements on TCAM.Under this background,this paper designs and implements a TCAM system that can automatically adjust the voltage swing of match-line.This paper first briefly introduces the structure and basic principles of TCAM,analyzes the current research on reducing power consumption and improving search speed of TCAM circuits,and summarizes the related ideas and methods for reducing the power consumption of match-line.Afterwards,based on the idea of reducing the voltage swing of the matching line,this paper designs a simple structure of the match-line voltage detector circuit.In theory,the circuit can work on the voltage swing of the match-line close to zero.The lower voltage swing of the matching line effectively saves power consumption of the circuit and also improves the search speed.In order to ensure robustness of proposed circuit,this paper also designed a negative feedback adjustment mechanism to adjust the match-line voltage swing.When a circuit error is detected,the timing control circuit can automatically adjust the voltage swing of match-line,thereby correcting the circuit error and improving the robustness of proposed system.Next,through the simulation analysis of the circuit,the paper verifies the function of the proposed system,and summarizes the mathematical relationship between the circuit search time and the voltage swing of match-line.Finally,based on the proposed scheme,the paper uses a standard CMOS 55 nm 1.2V process library to implement the proposed TCAM with a capacity of 64 words × 144bits(T-bit).The post layout simulation results show that proposed TCAM achieves0.18 f J/bit/search for the ML power dissipation with less than 750 ps search time.Normalized ML Energy-Delay-Product(EDP)achieves 0.14 f J?ns/bit/search which accounts for 16.1%and 53.8% that of [8] and [31] separately.